Application of CNNs in Home Security

Abstract

This study uses a deep learning model to detect the existence of a flame in an unused kitchen burner. This can also be used to sound an alarm or warn people who need to know. In this paper, a deep learning model is developed using transfer learning from the well-known Inceptionv3 Convolutional Neural Network (CNN) model. After that, the results are compared to a model created using the ResNet50 pretrained model. Originally, the Inceptionv3 and ResNet50 models were trained to distinguish up to 1000 image classes. Flame-On and Flame-Off are two image classes used in the proposed work. A total of 276 photos from two image classes make up the training dataset. Data augmentation including flipping, scaling, and rotation is used to increase the diversity of training data. When tested for 50 test images, the inceptionv3-based model surpasses the ResNet50-based model with a validation accuracy of 98 percent and a test accuracy of 94 percent. The proposed models are built and trained using the Matlab Deep Network Designer tool with predetermined training options.